Research map / Multi-agent systems and games / Cooperative multi-agent RL
Multi-Agent Incentive Communication via Decentralized Teammate Modeling
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use MAIC's teammate-modeled incentive communication instead of QMIX or communication baselines | Cooperative MARL · Level-Based Foraging | test return | Helped |
| Use MAIC instead of QMIX, NDQ, or TMC | Cooperative MARL · Hallway | test win rate | Helped |
| Use MAIC instead of communication and value-factorization baselines | QMIX-based MARL · SMAC, 13 scenarios | average test win rate | Helped |
| Integrate MAIC communication with VDN, QMIX, or QPLEX | Value decomposition methods · SMAC: MMM2 and 2c vs 64zg | test win rate | Helped |
| Remove sparse communication control and use uniform weights (MAIC-FC) (ablation) | MAIC · SMAC: MMM2 and 2c vs 64zg | test win rate | Hurt |
| Remove teammate modeling and broadcast non-tailored messages (MAIC-NC) (ablation) | MAIC · SMAC: MMM2 and 2c vs 64zg | test win rate | Hurt |
| Increase QMIX network size to approximately MAIC's parameter count (ablation) | QMIX · SMAC: MMM2 and 2c vs 64zg | test win rate | No clear effect |
| Prune 90% of messages | MAIC · MMM2 | test win rate: 96.7 ± 0.50 -> 96.5 ± 0.55 | No clear effect |
| Prune 90% of messages | TMC · MMM2 | test win rate: 85.6 ± 5.45 -> 78.7 ± 5.00 | Hurt |
| Prune 90% of messages | NDQ · MMM2 | test win rate: 72.5 ± 6.37 -> 7.50 ± 4.24 | Hurt |
| Prune 90% of messages | VBC · MMM2 | test win rate: 59.8 ± 1.05 -> 57.8 ± 2.23 | Hurt |
| Use incentive messages to bias teammates' value functions rather than augmenting local policy inputs with exchanged observations or embeddings | Cooperative MARL · SMAC super hard maps | test win rate | Helped |
Closest papers by meaning
- Efficient Communication in Multi-Agent Reinforcement Learning with Implicit Consensus Generation (AAAI 2025 · similarity 0.91)
- Multi-Agent Reinforcement Learning with Communication-Constrained Priors (NeurIPS 2025 · similarity 0.90)
- Social Influence as Intrinsic Motivation for Multi-Agent Deep Reinforcement Learning (ICML 2019 · similarity 0.90)
- Coordination Between Individual Agents in Multi-Agent Reinforcement Learning (AAAI 2021 · similarity 0.90)
- Bridging Training and Execution via Dynamic Directed Graph-Based Communication in Cooperative Multi-Agent Systems (AAAI 2025 · similarity 0.89)
- Language Grounded Multi-agent Reinforcement Learning with Human-interpretable Communication (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.